TheoremBase

Tensoring with a one-point measure space turns the coordinate Tonelli theorem into an iterated integral for the indicator of the set; the inner integral is the one-dimensional measure of a line section, which vanishes by hypothesis.

Proof

We use the notation of the statement.

The measure-theoretic set-up. Let Z={z}Z=\{z\} be a one-point set and let G={∅,Z}\mathcal{G}=\{\varnothing,Z\} and δ\delta be as in claim 3 of Assembly of Measure Spaces: Restriction, Transport, One-Point Spaces, and Countable Disjoint Unions, so that G\mathcal{G} is a σ\sigma-algebra on ZZ, δ(∅)=0\delta(\varnothing)=0 and δ(Z)=1\delta(Z)=1 define a measure on it, and ∫Zg dδ=g(z)\int_{Z}g\,d\delta=g(z) for every g:Z→[0,∞]g:Z\to[0,\infty]. The constant sequence with every term ZZ has union ZZ and δ(Z)=1<∞\delta(Z)=1<\infty, so δ\delta is σ\sigma-finite.

We apply Finite Products of Lebesgue Measure and Coordinate Integration on Rl\mathbb{R}^l with l=nl=n, with the measure space (Z,G,δ)(Z,\mathcal{G},\delta) in the role of (Y,G,μ)(Y,\mathcal{G},\mu), and with the given index ii. By Lebesgue Measure on Rn\mathbb{R}^n the σ\sigma-algebra Bn\mathcal{B}_{n} of that lemma is B(Rn)\mathcal{B}(\mathbb{R}^{n}) and the measure λn\lambda_{n} of that lemma is Lebesgue measure on it, so no distinction between the two notations is needed. When n=1n=1 we use the conventions recorded in that lemma for l=1l=1, by which Rl−1×Y\mathbb{R}^{l-1}\times Y denotes YY, λl−1⊗μ\lambda_{l-1}\otimes\mu denotes μ\mu, a pair (θ′,y)(\theta',y) denotes yy, and Ψ1(t,y)=(t,y)\Psi_{1}(t,y)=(t,y).

The integral of the indicator. The set N×ZN\times Z is a product of a member of B(Rn)\mathcal{B}(\mathbb{R}^{n}) with a member of G\mathcal{G}, hence belongs to the product σ\sigma-algebra B(Rn)⊗G\mathcal{B}(\mathbb{R}^{n})\otimes\mathcal{G}. By The Integral of an Indicator Function is the Measure of the Set its indicator function 1N×Z\mathbf{1}_{N\times Z} is a nonnegative measurable function on Rn×Z\mathbb{R}^{n}\times Z with

∫Rn×Z1N×Z d(λn⊗δ)=(λn⊗δ)(N×Z)=λn(N) δ(Z)=λn(N),\int_{\mathbb{R}^{n}\times Z}\mathbf{1}_{N\times Z}\,d(\lambda_{n}\otimes\delta)=(\lambda_{n}\otimes\delta)(N\times Z)=\lambda_{n}(N)\,\delta(Z)=\lambda_{n}(N),

the middle equality by Existence and Uniqueness of the Product Measure, which characterises λn⊗δ\lambda_{n}\otimes\delta by its values on such products.

By claim 3 of Finite Products of Lebesgue Measure and Coordinate Integration on Rl\mathbb{R}^l,

λn(N)=∫Rn−1×Z(∫R1N×Z(Ψi(t,(θ′,z))) dλ(t)) d(λn−1⊗δ)(θ′,z),\lambda_{n}(N)=\int_{\mathbb{R}^{n-1}\times Z}\Bigl(\int_{\mathbb{R}}\mathbf{1}_{N\times Z}\bigl(\Psi_{i}(t,(\theta',z))\bigr)\,d\lambda(t)\Bigr)\,d(\lambda_{n-1}\otimes\delta)(\theta',z),

and for every (θ′,z)(\theta',z) the inner integrand is a B(R)\mathcal{B}(\mathbb{R})-measurable function of tt.

The inner integral vanishes. Fix (θ′,z)(\theta',z) with θ′=(θ1,…,θn−1)∈Rn−1\theta'=(\theta_{1},\dots,\theta_{n-1})\in\mathbb{R}^{n-1}. By the description of the insertion map in claim 2 of Finite Products of Lebesgue Measure and Coordinate Integration on Rl\mathbb{R}^l,

Ψi(t,(θ′,z))=((θ1,…,θi−1,t,θi,…,θn−1), z),\Psi_{i}\bigl(t,(\theta',z)\bigr)=\bigl((\theta_{1},\dots,\theta_{i-1},t,\theta_{i},\dots,\theta_{n-1}),\,z\bigr),

so 1N×Z(Ψi(t,(θ′,z)))=1T(t)\mathbf{1}_{N\times Z}(\Psi_{i}(t,(\theta',z)))=\mathbf{1}_{T}(t), where

T={ t∈R  :  (θ1,…,θi−1,t,θi,…,θn−1)∈N }.T=\{\,t\in\mathbb{R}\;:\;(\theta_{1},\dots,\theta_{i-1},t,\theta_{i},\dots,\theta_{n-1})\in N\,\}.

Put y0=(θ1,…,θi−1,0,θi,…,θn−1)y_{0}=(\theta_{1},\dots,\theta_{i-1},0,\theta_{i},\dots,\theta_{n-1}). For t∈Rt\in\mathbb{R} the point y0+teiy_{0}+te_{i} has iith coordinate 0+t=t0+t=t and, for p≠ip\neq i, the same ppth coordinate as y0y_{0}, because the corresponding coordinate of eie_{i} is 00; hence y0+tei=(θ1,…,θi−1,t,θi,…,θn−1)y_{0}+te_{i}=(\theta_{1},\dots,\theta_{i-1},t,\theta_{i},\dots,\theta_{n-1}) and

T={ t∈R  :  y0+tei∈N }.T=\{\,t\in\mathbb{R}\;:\;y_{0}+te_{i}\in N\,\}.

By hypothesis TT is λ1\lambda_{1}-null, so by Null Set of a Measure there is a Borel set B⊆RB\subseteq\mathbb{R} with T⊆BT\subseteq B and λ(B)=0\lambda(B)=0. The function 1T\mathbf{1}_{T} is B(R)\mathcal{B}(\mathbb{R})-measurable by the previous paragraph, and TT is the set where it takes the value 11, so TT is a Borel subset of R\mathbb{R}; hence claim 2 of Basic Properties of a Measure gives λ(T)≤λ(B)=0\lambda(T)\le\lambda(B)=0. By The Integral of an Indicator Function is the Measure of the Set,

∫R1T dλ=λ(T)=0.\int_{\mathbb{R}}\mathbf{1}_{T}\,d\lambda=\lambda(T)=0 .

Conclusion. The inner integral vanishes for every (θ′,z)(\theta',z), so the outer integrand is the function taking the value 00 everywhere on Rn−1×Z\mathbb{R}^{n-1}\times Z; that function is 1∅\mathbf{1}_{\varnothing}, so by The Integral of an Indicator Function is the Measure of the Set its integral is (λn−1⊗δ)(∅)=0(\lambda_{n-1}\otimes\delta)(\varnothing)=0. Therefore λn(N)=0\lambda_{n}(N)=0. Since N∈B(Rn)N\in\mathcal{B}(\mathbb{R}^{n}) and λn(N)=0\lambda_{n}(N)=0, taking B=NB=N in Null Set of a Measure shows that NN is null.

Citations

Loading…

Dependencies

Uses0

Loading…

Comments

Log in to comment.

Loading…